Dataloader cannot use num_workers>0L

[cregouby]

as stated in #37 , Dataloader fail to run with num_workers>0L

Current behavior

When running with

tabnet/R/model.R

Line 110 in a16b4cc

#' @param num_workers (int, optional): how many subprocesses to use for data

, we get

# Error : callr subprocess failed: external pointer is not valid
# Run `rlang::last_error()` to see where the error occurred.
# De plus : Messages d'avis :
# 1: Datasets used with parallel dataloader (num_workers > 0) shouldn't have fields containing tensors as they can't be correctly passed to the wroker subprocesses.

Impact

The dataloader is currently a bottleneck in situation where the torch engine ( GPU ) is faster to process tensors than the dataloader ( single CPU core) to prepare it.